AMDGPU: Clear kill flags after FoldZeroHighBits

[changpeng]

After folding, all uses of the result register are going to be replaced by the operand register. The kill flags on the uses of the result and operand registers are no longer valid after the replacement, and need to be cleared.
The only exception is, however, if the kill flag is set for the operand register, we are sure the last use of the result register is the new last use of the operand register, and thus we are safe to keep the kill flags.

[llvmbot]

@llvm/pr-subscribers-backend-amdgpu

Author: Changpeng Fang (changpeng)

Changes

After the folding, all the uses of the result register are going to be replaced by the operand register. The kill flags on the uses of the result register are no longer valid after the replacement.
The only exception is, however, if the kill flag is set for the operand register, we are sure the last use of the result register is the new last use of the operand register, and thus we are safe to keep the kill flags.

---

Full diff: https://github.com/llvm/llvm-project/pull/99582.diff

2 Files Affected:

• (modified) llvm/lib/Target/AMDGPU/SIFoldOperands.cpp (+2)
• (added) llvm/test/CodeGen/AMDGPU/fold-zero-high-bits-clear-kill-flags.mir (+54)

diff --git a/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp b/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp
index 0e8c96625b221..a1524a2c051f7 100644
--- a/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp
+++ b/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp
@@ -1361,6 +1361,8 @@ bool SIFoldOperands::tryFoldZeroHighBits(MachineInstr &MI) const {
     return false;
 
   Register Dst = MI.getOperand(0).getReg();
+  if (!MI.getOperand(2).isKill())
+    MRI->clearKillFlags(Dst);
   MRI->replaceRegWith(Dst, SrcDef->getOperand(0).getReg());
   MI.eraseFromParent();
   return true;
diff --git a/llvm/test/CodeGen/AMDGPU/fold-zero-high-bits-clear-kill-flags.mir b/llvm/test/CodeGen/AMDGPU/fold-zero-high-bits-clear-kill-flags.mir
new file mode 100644
index 0000000000000..baaca76bfd8a8
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/fold-zero-high-bits-clear-kill-flags.mir
@@ -0,0 +1,54 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
+# RUN: llc -mtriple=amdgcn--amdhsa -mcpu=gfx940 -verify-machineinstrs -run-pass si-fold-operands -o - %s | FileCheck -enable-var-scope -check-prefix=GCN %s
+
+---
+name:            fold_zero_high_bits_src1_alive
+tracksRegLiveness: true
+
+body:             |
+  bb.0:
+    liveins: $vgpr0, $vgpr1
+    ; GCN-LABEL: name: fold_zero_high_bits_src1_alive
+    ; GCN: liveins: $vgpr0, $vgpr1
+    ; GCN-NEXT: {{  $}}
+    ; GCN-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+    ; GCN-NEXT: [[V_ADD_U16_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U16_e64 [[COPY]], 1, 0, implicit $exec
+    ; GCN-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 65535
+    ; GCN-NEXT: [[V_MUL_U32_U24_e64_:%[0-9]+]]:vgpr_32 = V_MUL_U32_U24_e64 [[V_ADD_U16_e64_]], 1, 0, implicit $exec
+    ; GCN-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+    ; GCN-NEXT: [[V_SUB_U16_e64_:%[0-9]+]]:vgpr_32 = V_SUB_U16_e64 [[V_ADD_U16_e64_]], [[COPY1]], 0, implicit $exec
+    %0:vgpr_32 = COPY $vgpr0
+    %1:sreg_32 = S_MOV_B32 1
+    %2:vgpr_32 = V_ADD_U16_e64 %0:vgpr_32, %1:sreg_32, 0, implicit $exec
+    %3:sreg_32 = S_MOV_B32 65535
+    %4:vgpr_32 = V_AND_B32_e64 %3:sreg_32, %2:vgpr_32, implicit $exec
+    %5:vgpr_32 = V_MUL_U32_U24_e64 killed %4:vgpr_32, %1:sreg_32, 0, implicit $exec
+    %6:vgpr_32 = COPY $vgpr1
+    %7:vgpr_32 = V_SUB_U16_e64 %2:vgpr_32, %6:vgpr_32, 0, implicit $exec
+...
+
+---
+name:            fold_zero_high_bits_src1_killed
+tracksRegLiveness: true
+
+body:             |
+  bb.0:
+    liveins: $vgpr0, $vgpr1
+    ; GCN-LABEL: name: fold_zero_high_bits_src1_killed
+    ; GCN: liveins: $vgpr0, $vgpr1
+    ; GCN-NEXT: {{  $}}
+    ; GCN-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+    ; GCN-NEXT: [[V_ADD_U16_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U16_e64 [[COPY]], 1, 0, implicit $exec
+    ; GCN-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+    ; GCN-NEXT: [[V_SUB_U16_e64_:%[0-9]+]]:vgpr_32 = V_SUB_U16_e64 [[V_ADD_U16_e64_]], [[COPY1]], 0, implicit $exec
+    ; GCN-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 65535
+    ; GCN-NEXT: [[V_MUL_U32_U24_e64_:%[0-9]+]]:vgpr_32 = V_MUL_U32_U24_e64 killed [[V_ADD_U16_e64_]], 1, 0, implicit $exec
+    %0:vgpr_32 = COPY $vgpr0
+    %1:sreg_32 = S_MOV_B32 1
+    %2:vgpr_32 = V_ADD_U16_e64 %0:vgpr_32, %1:sreg_32, 0, implicit $exec
+    %6:vgpr_32 = COPY $vgpr1
+    %7:vgpr_32 = V_SUB_U16_e64 %2:vgpr_32, %6:vgpr_32, 0, implicit $exec
+    %3:sreg_32 = S_MOV_B32 65535
+    %4:vgpr_32 = V_AND_B32_e64 %3:sreg_32, killed %2:vgpr_32, implicit $exec
+    %5:vgpr_32 = V_MUL_U32_U24_e64 killed %4:vgpr_32, %1:sreg_32, 0, implicit $exec
+...